There are few slight differences in the execution paths, nothing jumps out (it *looks* like the authorization to write to the CF is checked for each statement in the batch, not sure how heavy that is.).
If you send a batch with more statements that concurrent_writers in the yaml some of those statements will have to wait for an available writer before completing. This will introduce some latency to the query. You can check pending tasks using nodetool tpstats. Before we get into it further some thoughts: * what cassandra version ? * you are running the tests one after another on the same running cassandra process ? Or are you running it against a new process. * have a look at the nodetool cfstats to see the write latency for the cf, this is the latency for the actual write. Does it change ? * Use jconsole to look at the o.a.c.db.StorageProxy MBean, the latency there is for entire requests. * perhaps take a look at the stress testing tools in the distribution and see if their results concur with yours. If you are still having problems let us know and include the python script. Cheers ----------------- Aaron Morton Freelance Developer @aaronmorton http://www.thelastpickle.com On 25/01/2012, at 3:33 PM, Blake Visin wrote: > So I decided that it would be beneficial to use batching in my application > since I am doing many, many inserts. When I implemented batching in CQL > using 'BEGIN BATCH'..'APPLY BATCH' I saw a significant decrease in the speed > of inserts, no matter the number of insert statements I included between > begin and apply. I created a simple benchmark script in Python and posted > the results here: > > https://docs.google.com/spreadsheet/ccc?key=0AoNsUtyQ2cAodGpxLV9MODNJV3BFMzBicUhnUjM2M0E > > As you can see, the larger I made the batches, the longer they took. > > Any ideas where to go from here? > > Thanks, > Blake
